Merge pull request spmallick#486 from vkoriukina/vk/face_detection_comparision_dnn_gpu_update

FaceDetectionComparison: added DNN GPU usage

Author: vikasgupta-github

FaceDetectionComparison/CMakeLists.txt

@@ -1,11 +1,18 @@
 cmake_minimum_required(VERSION 2.8.12)
+set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
 
 PROJECT(face_detection)
 
 find_package( OpenCV REQUIRED )
 
-include_directories( ${OpenCV_INCLUDE_DIRS})
+message(STATUS "OpenCV library status:")
+message(STATUS "    config: ${OpenCV_DIR}")
+message(STATUS "    version: ${OpenCV_VERSION}")
+message(STATUS "    libraries: ${OpenCV_LIBS}")
+message(STATUS "    include path: ${OpenCV_INCLUDE_DIRS}")
 
+include_directories( ${OpenCV_INCLUDE_DIRS})
 include(./dlib/dlib/cmake)
 
 MACRO(add_example name)

FaceDetectionComparison/README.md

@@ -1,42 +1,130 @@
-There are 10 code files in this repo - 5 for C++ and 5 for Python
+There are 10 code files in this repo - 5 for C++ and 5 for Python:
+
 1. face_detection_opencv_haar.cpp and face_detection_opencv_haar.py - For Haar based face detection
 2. face_detection_opencv_dnn.cpp and face_detection_opencv_dnn.py - For OpenCV DNN based face detection
 3. face_detection_dlib_hog.cpp and face_detection_dlib_hog.py - for dlib hog based face detection
 4. face_detection_dlib_mmod.cpp and face_detection_dlib_mmod.py - for dlib mmod based face detection
 5. run-all.cpp and run-all.py - for running all the 4 together
 
-
 First of all Unzip the dlib.zip file
 
 ## For C++
+
 **Compile**
-cmake .
+
+Add path to the properly build OpenCV with DNN GPU Support and your CUDA:
+
+```
+cmake -D OpenCV_DIR=~/opencv -D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda/ .
 make
+```
+
+## For Python
+
+_Note_: Our code is tested using Python 3.7.5, but it should also work with any other python3.x.
+
+Install virtualenv:
+
+```
+pip install virtualenv
+```
+
+Create new virtual environment `env` and activate it:
+
+```
+python -m venv env
+source  env/bin/activate
+```
+
+Install numpy:
+
+```
+pip install numpy
+```
+
+Install dlib:
+
+```
+pip install dlib
+```
+
+Create symlink to the properly build OpenCV with DNN GPU Support:
+
+```
+cd env/lib/python3.x/site-packages/
+ln -s ~/opencv/build/cv2.so cv2.so
+```
 
 ## Run
-**If you dont give any filename, it will use the webcam**
+
+**If you don't pass any filename, it will use the web cam**
 
 ### For individual face detectors
+
 **C++**
+
+```
 ./face_detection_XXXX <filename>
+```
+
+_Note:_ for `face_detection_opencv_dnn.cpp` you can pass up to 3 arguments:
+
+- video filename, if you'd like to run inference on a video instead of a camera:
+
+```
+./face_detection_opencv_dnn.out <filename>
+```
+
+- device, if you want to use CPU instead of CPU:
+
+```
+./face_detection_opencv_dnn.out "" cpu
+```
+
+- framework to specify Caffe (caffe) or TensorFlow (tf) network to use. Caffe network is set by default:
+
+```
+./face_detection_opencv_dnn.out "" gpu tf
+```
 
 **Python**
-python face_detection_XXXX.py <filename>
+
+```
+python face_detection_XXXX.py -video <filename>
+```
+
+_Note:_ for `face_detection_opencv_dnn.py` you can pass up to 3 arguments:
+
+- filename, if you'd like to run inference on a video instead of a camera:
+
+```
+python face_detection_opencv_dnn.out --video <filename>
+```
+
+- device, if you want to use CPU instead of GPU:
+
+```
+python face_detection_opencv_dnn.out --video <filename> --device cpu
+```
+
+- framework to specify Caffe (caffe) or TensorFlow (tf) network to use. Caffe network is set by default:
+
+```
+python face_detection_opencv_dnn.out --video <filename> --device cpu --framework tf
+```
 
 ### For running all together
-**C++**
-./run-all <filename>
 
-**Python**
-run-all.py <filename>
+**C++** ./run-all <filename>
 
+**Python** python run-all.py --video <filename>
 
 # AI Courses by OpenCV
 
-Want to become an expert in AI? [AI Courses by OpenCV](https://opencv.org/courses/) is a great place to start. 
+Want to become an expert in AI? [AI Courses by OpenCV](https://opencv.org/courses/) is a great place to start.
 
 <a href="https://opencv.org/courses/">
-<p align="center"> 
+<p align="center">
 <img src="https://www.learnopencv.com/wp-content/uploads/2020/04/AI-Courses-By-OpenCV-Github.png">
 </p>
 </a>

FaceDetectionComparison/dlib.zip

@@ -36,44 +36,49 @@ void detectFaceDlibHog(frontal_face_detector hogFaceDetector, Mat &frameDlibHog,
 
     for ( size_t i = 0; i < faceRects.size(); i++ )
     {
-      int x1 = (int)(faceRects[i].left() * scaleWidth);
-      int y1 = (int)(faceRects[i].top() * scaleHeight);
-      int x2 = (int)(faceRects[i].right() * scaleWidth);
-      int y2 = (int)(faceRects[i].bottom() * scaleHeight);
-      cv::rectangle(frameDlibHog, Point(x1, y1), Point(x2, y2), Scalar(0,255,0), (int)(frameHeight/150.0), 4);
+        int x1 = (int)(faceRects[i].left() * scaleWidth);
+        int y1 = (int)(faceRects[i].top() * scaleHeight);
+        int x2 = (int)(faceRects[i].right() * scaleWidth);
+        int y2 = (int)(faceRects[i].bottom() * scaleHeight);
+        cv::rectangle(frameDlibHog, Point(x1, y1), Point(x2, y2), Scalar(0,255,0), (int)(frameHeight/150.0), 4);
     }
 }
 
+
 int main( int argc, const char** argv )
 {
-  frontal_face_detector hogFaceDetector = get_frontal_face_detector();
-
-  VideoCapture source;
-  if (argc == 1)
-      source.open(0);
-  else
-      source.open(argv[1]);
-  Mat frame;
-
-  double tt_dlibHog = 0;
-  double fpsDlibHog = 0;
-  while(1)
-  {
-      source >> frame;
-      if(frame.empty())
-          break;
-
-      double t = cv::getTickCount();
-      detectFaceDlibHog ( hogFaceDetector, frame );
-      tt_dlibHog = ((double)cv::getTickCount() - t)/cv::getTickFrequency();
-      fpsDlibHog = 1/tt_dlibHog;
-      putText(frame, format("DLIB HoG ; FPS = %.2f",fpsDlibHog), Point(10, 50), FONT_HERSHEY_SIMPLEX, 1.4, Scalar(0, 0, 255), 4);
-      imshow( "DLIB - HoG Face Detection", frame );
-      int k = waitKey(5);
-      if(k == 27)
-      {
-        destroyAllWindows();
-        break;
+    frontal_face_detector hogFaceDetector = get_frontal_face_detector();
+
+    VideoCapture source;
+    if (argc == 1)
+        source.open(0, CAP_V4L);
+    else
+        source.open(argv[1]);
+
+    Mat frame;
+
+    double tt_dlibHog = 0;
+    double fpsDlibHog = 0;
+
+    while (true)
+    {
+        source >> frame;
+        if (frame.empty())
+            break;
+
+        double t = cv::getTickCount();
+        detectFaceDlibHog(hogFaceDetector, frame);
+        tt_dlibHog = ((double)cv::getTickCount() - t)/cv::getTickFrequency();
+        fpsDlibHog = 1/tt_dlibHog;
+
+        putText(frame, format("DLIB HoG; FPS = %.2f",fpsDlibHog), Point(10, 50), FONT_HERSHEY_SIMPLEX, 1.3, Scalar(0, 0, 255), 4);
+        imshow("DLIB - HoG Face Detection", frame);
+
+        int k = waitKey(5);
+        if(k == 27)
+        {
+            destroyAllWindows();
+            break;
+        }
       }
-    }
 }

FaceDetectionComparison/face_detection_dlib_hog.cpp

@@ -1,16 +1,18 @@
-from __future__ import division
+import argparse
+import os
+import time
+
 import cv2
 import dlib
-import time
-import sys
+
 
 def detectFaceDlibHog(detector, frame, inHeight=300, inWidth=0):
 
     frameDlibHog = frame.copy()
     frameHeight = frameDlibHog.shape[0]
     frameWidth = frameDlibHog.shape[1]
     if not inWidth:
-        inWidth = int((frameWidth / frameHeight)*inHeight)
+        inWidth = int((frameWidth / frameHeight) * inHeight)
 
     scaleHeight = frameHeight / inHeight
     scaleWidth = frameWidth / inWidth
@@ -23,48 +25,89 @@ def detectFaceDlibHog(detector, frame, inHeight=300, inWidth=0):
     bboxes = []
     for faceRect in faceRects:
 
-        cvRect = [int(faceRect.left()*scaleWidth), int(faceRect.top()*scaleHeight),
-                  int(faceRect.right()*scaleWidth), int(faceRect.bottom()*scaleHeight) ]
+        cvRect = [
+            int(faceRect.left() * scaleWidth),
+            int(faceRect.top() * scaleHeight),
+            int(faceRect.right() * scaleWidth),
+            int(faceRect.bottom() * scaleHeight),
+        ]
         bboxes.append(cvRect)
-        cv2.rectangle(frameDlibHog, (cvRect[0], cvRect[1]), (cvRect[2], cvRect[3]), (0, 255, 0), int(round(frameHeight/150)), 4)
+        cv2.rectangle(
+            frameDlibHog,
+            (cvRect[0], cvRect[1]),
+            (cvRect[2], cvRect[3]),
+            (0, 255, 0),
+            int(round(frameHeight / 150)),
+            4,
+        )
     return frameDlibHog, bboxes
 
-if __name__ == "__main__" :
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser(description="Face detection")
+    parser.add_argument("--video", type=str, default="", help="Path to video file")
+    args = parser.parse_args()
+
+    source = args.video
     hogFaceDetector = dlib.get_frontal_face_detector()
 
-    source = 0
-    if len(sys.argv) > 1:
-        source = sys.argv[1]
+    outputFolder = "output-hog-videos"
+    if not os.path.exists(outputFolder):
+        os.makedirs(outputFolder)
 
-    cap = cv2.VideoCapture(source)
+    if source:
+        cap = cv2.VideoCapture(source)
+        outputFile = os.path.basename(source)[:-4] + ".avi"
+    else:
+        cap = cv2.VideoCapture(0, cv2.CAP_V4L)
+        outputFile = "grabbed_from_camera.avi"
     hasFrame, frame = cap.read()
 
-    vid_writer = cv2.VideoWriter('output-hog-{}.avi'.format(str(source).split(".")[0]),cv2.VideoWriter_fourcc('M','J','P','G'), 15, (frame.shape[1],frame.shape[0]))
+    vid_writer = cv2.VideoWriter(
+        os.path.join(outputFolder, outputFile),
+        cv2.VideoWriter_fourcc("M", "J", "P", "G"),
+        15,
+        (frame.shape[1], frame.shape[0]),
+    )
 
     frame_count = 0
     tt_dlibHog = 0
-    while(1):
+
+    while True:
         hasFrame, frame = cap.read()
         if not hasFrame:
             break
-        frame_count += 1
 
+        frame_count += 1
         t = time.time()
-        outDlibHog, bboxes = detectFaceDlibHog(hogFaceDetector,frame)
+
+        outDlibHog, bboxes = detectFaceDlibHog(hogFaceDetector, frame)
         tt_dlibHog += time.time() - t
         fpsDlibHog = frame_count / tt_dlibHog
 
-        label = "DLIB HoG ; ; FPS : {:.2f}".format(fpsDlibHog)
-        cv2.putText(outDlibHog, label, (10, 50), cv2.FONT_HERSHEY_SIMPLEX, 1.4, (0, 0, 255), 3, cv2.LINE_AA)
+        label = "DLIB HoG; FPS : {:.2f}".format(fpsDlibHog)
+        cv2.putText(
+            outDlibHog,
+            label,
+            (10, 50),
+            cv2.FONT_HERSHEY_SIMPLEX,
+            1.3,
+            (0, 0, 255),
+            3,
+            cv2.LINE_AA,
+        )
 
         cv2.imshow("Face Detection Comparison", outDlibHog)
 
         vid_writer.write(outDlibHog)
+
         if frame_count == 1:
             tt_dlibHog = 0
 
-        k = cv2.waitKey(10)
+        k = cv2.waitKey(5)
         if k == 27:
             break
+
     cv2.destroyAllWindows()
     vid_writer.release()